A while ago we wrote about why most AI writing reads the same. The short version: every model has a center of gravity, a safe and tidy way of writing that it drifts back to unless something strong pulls it elsewhere. That post was the theory. This one is the reckoning.
Because here is the uncomfortable thing about running a product that writes every day: the sameness problem does not only live in other people's tools. It can creep into yours. And the only way to know is to look. So we looked.
We audited ourselves, and it stung
We took a two-week sample of the posts bbuddy had drafted and read every one of them the way a tough editor would. Not "is this fine". More like "would a real person, scrolling their own feed, feel like the same voice was talking at them every single day".
The answer, often, was yes. Individually the posts were good. Read back to back, they blurred. More than half leaned on the exact same rhetorical move, the little "it is not this, it is that" flip. Almost none of them cited a concrete number. Not one anchored on a real thing that had actually happened that week. And nearly every post, regardless of topic, followed the identical shape: a hook, a turn, a tidy little resolution.
None of that is a bug in the usual sense. Every draft passed. The problem was at the level of the feed, and a feed is exactly what your audience experiences. Nobody reads one of your posts. They read the drip of all of them.
Why variety is genuinely hard
The naive fix is to tell the writer "be more varied". It does not work, and it is worth being honest about why. When you ask a model to vary itself, it reaches for its own idea of variety, which is another average. You get the same three tricks in a different order. The instruction to be surprising is not surprising.
Real variety does not come from asking for it. It comes from changing what a post is built out of and remembering what you already made. That reframing is what the last stretch of work has been about, and it broke down into a few plain ideas.
What we changed
Anchor posts on something real. Instead of writing from abstractions, bbuddy now reaches for fresh, concrete material from your own world before it drafts: a post of yours that actually did well, a real number, something happening this week. A post built on a true detail cannot help but be more specific than one spun from thin air. The rule we hold it to is simple: anchor on a real thing, never invent one.
Remember the shape of what you already posted. Repetition is not only about topic. It is about how a post opens, how it lands, its rhythm. bbuddy now keeps the shapes of your recent posts in view while it writes the next one, and treats "you already opened three posts this way" as a reason to do something else. Your themes can repeat. Repeating your themes is your identity. Repeating your moves is a tic.
Draw from a wide menu of shapes. There are far more ways to write a post than the tidy little essay. A blunt observation. A real question. A one-liner. A short story. A number that speaks for itself. We gave bbuddy a deep catalog of these and have it rotate through them deliberately, so a corporate account never gets handed a confessional and a personal account is not forced into a stiff format. The essay is now one shape among many, not the default everything collapses toward.
Treat every channel as its own conversation. This one came a little earlier and taught us a lot. A low-volume channel, say one post a day, was opening with the same flagship line for days on end, while a busy channel stayed varied. The writer simply could not see the quiet channel's own history clearly enough. Now each platform is its own running thread. The goal on each one is to move the conversation forward, not restate yesterday's headline in fresh paint.
The principle underneath all of it
Everything above is one idea wearing four coats. You do not get variety by prescribing it. You get it by feeding the writer richer, truer, more varied material and by letting it remember what it already said. Prescribe the output and you narrow it. Ground the inputs and you widen it. We keep relearning this every time we are tempted to just add another instruction.
It matters that all of this is built in, not something you have to babysit. You should not have to prompt your way out of sameness every morning. The point of an agent is that it carries that standard for you.
Where this still falls short
A young brand with a thin history is harder to keep varied than an established one, simply because there is less real material to anchor on. Variety compounds with use: the more of your world bbuddy has read, the less it has to fall back on its own defaults. And "different" is not automatically "good". A post can be fresh in shape and still not land. Chasing novelty for its own sake is its own trap, and we watch for it.
But the direction is set, and we are confident about it. Sameness is the enemy. Not because variety is a nice garnish, but because a brand that sounds like a template is invisible, and invisible is the one thing content cannot afford to be.
If you want the theory behind all this, it lives in why most AI writing reads the same, and the bigger picture of where bbuddy is heading is in the agentic marketing manifesto.